Mechanistic elucidation guided by covalent inhibitors for the development of anti-diabetic PPARγ ligands† †Electronic supplementary information (ESI) available: Supplementary figures and tables, experimental methods, procedures for synthesis, and full characterization data of compounds. See DOI: 10.

We revealed the X-ray structure of PPARγ co-crystallized with SR1664 bound to the alternate binding site of PPARγ and confirmed that this blocks the phosphorylation of Ser273.


Introduction
Insulin resistance, a major symptom of type II diabetes, is a condition in which body cells become resistant to the normal actions of insulin. 1 Because of the potent insulin-sensitizing effects of peroxisome proliferator-activated receptor gamma (PPARg)-targeting drugs such as rosiglitazone and pioglitazone, PPARg has been considered to be a major therapeutic target for the treatment of type II diabetes, although the underlying molecular mechanisms are still unclear. [2][3][4] However, Choi et al. proposed a plausible mechanism by which anti-diabetic PPARg ligands affect insulin sensitivity. 5 They demonstrated that the obesity-induced phosphorylation of PPARg at serine 273 (Ser273) results in the dysregulation of a subset of genes involved in insulin resistance and revealed that anti-diabetic PPARg ligands effectively block this phosphorylation. 5 Based on these ndings, it was suggested that the efficacy of anti-diabetic PPARg ligands is attributed to their ability to inhibit the phosphorylation of PPARg. Moreover, it was recently elucidated that an extracellular signal-regulated kinase (ERK)/cyclin-dependent kinase 5 (Cdk5) axis regulates the diabetogenic actions of PPARg through the phosphorylation of Ser273, 6 and that Ser273-phosphorylated PPARg is selectively recognized by thyroid hormone receptor-associated protein 3 (Thrap3) and regulates a diabetic gene set through these signaling pathways. 7 Although glitazones have remarkable effects in the treatment of type II diabetes, their use has declined because of their serious adverse effects including weight gain, uid retention, and congestive heart failure. 8,9 Glitazones both inhibit the phosphorylation of Ser273 and fully activate the expression of PPARg target genes, which is referred to as classical transcriptional agonism. 5 Full classical agonism has been suspected of causing serious side effects; thus, there have been consistent efforts to develop a selective PPARg modulator (SPPARgM) that exhibits reduced classical agonism while retaining potent effects on insulin sensitization. [10][11][12] Because PPARg has a large binding pocket and multiple interaction points with ligands, it is expected that PPARg activities can be selectively regulated through the site-specic binding of ligands. 13 The recent discovery of SR1664 as a representative SPPARgM helped demonstrate that the complete and selective modulation of PPARg activities is possible via specic ligand binding. Unlike glitazones, SR1664, which inhibits the phosphorylation of Ser273 without altering the transcriptional activity of PPARg, exerts potent in vivo anti-diabetic effects without causing uid retention and weight gain. 14 In consideration of these results, there is no doubt that designing a selective inhibitor of PPARg phosphorylation can be a powerful strategy for the development of a novel anti-diabetic agent targeting PPARg. However, the exact structural mechanism by which anti-diabetic PPARg ligands block Ser273 phosphorylation has not yet been elucidated.

Results and discussion
Alternate site binding of SR1664 blocks phosphorylation of PPARg at Ser273 To understand the exact molecular mechanism by which antidiabetic PPARg ligands can inhibit Ser273 phosphorylation, we resolved the crystal structure of the PPARg ligand-binding domain (LBD) complexed with SR1664 and the SRC1 coactivator peptide to a resolution of 2.20Å ( Fig. 1a and S1a †). As the molecular interaction between PPARg and SR1664 was mainly associated with the inhibition of Ser273 phosphorylation, we might gain more insight into the mechanism by which antidiabetic PPARg ligands block the phosphorylation of Ser273, by analyzing the co-crystal structures of the PPARg LBD and SR1664. SR1664 had a completely different binding mode from full PPARg agonists such as rosiglitazone which binds at the canonical binding pocket of PPARg via strong hydrogen bonding with helix 12 (Fig. 1b, blue). 16 In contrast, SR1664 bound to an alternate site which is dened as the region near the entrance of the canonical binding pocket occupied by the second MRL-20 molecule when two molar equivalents of an MRL-20 ligand are bound to PPARg (Fig. 1b, green). 17 To determine the phosphorylation-inhibiting effect of SR1664 bound at the alternate site, we investigated using GW9662, a synthetic irreversible PPARg inhibitor that covalently binds to cysteine 313 (Cys313 in PPARg2; Cys285 in PPARg1) on helix 3 (H3). 18 This covalent inhibitor completely blocks ligand engagement at the canonical binding pocket without fully inhibiting the alternate site binding of PPARg ligands. 17 In this study, GW9662 was used to block the canonical binding pocket to determine whether the alternate site binding of SR1664 affects coregulatory interactions. 17 However, based on the X-ray co-crystal structure, we found that SR1664 exhibited a steric clash with the phenyl group of GW9662 (Fig. S2 †), which can interfere in the binding event of SR1664 at the alternate site of PPARg when this ligand engages PPARg with the binding mode shown in the crystal structure.
Therefore, we designed and synthesized a smaller covalent inhibitor, SB1404 (1), by replacing the phenyl group with a methyl group (Fig. 1c). Compared to rosiglitazone or SR1664, SB1404 did not inhibit the Cdk5-mediated phosphorylation of PPARg at any concentration in vitro (Fig. 1d). Nevertheless, SB1404 completely blocked the inhibitory effect of rosiglitazone on PPARg phosphorylation, but it did not affect the inhibition of PPARg phosphorylation by SR1664 (Fig. 1e).
Furthermore, on the basis of the X-ray crystal structure of the PPARg LBD complexed with SB1404 and the SRC1 coactivator peptide (resolution, 2.80Å), we conrmed that SB1404 covalently bound to Cys313 on H3 and completely blocked the binding of rosiglitazone at the canonical binding pocket of PPARg ( Fig. 1f and S3 †). Unlike GW9662, SB1404 exhibited no steric clash with SR1664; thus, SR1664 can bind to SB1404labeled PPARg with the mode shown in the crystal structure ( Fig. 1g), indicating that alternate site binding of SR1664 directly inhibits PPARg phosphorylation. In fact, a different binding mode of SR1664 was recently reported, 15 but it is not possible for SR1664 to bind to SB1404-labeled PPARg through the reported binding mode (Fig. S4 †). Although more studies are necessary to determine which conformation of ligand binding is the major one, we clearly determined the functional effect of the alternate site binding of SR1664 in terms of phosphorylation inhibition at Ser273 by Cdk5. Site-specic binding of PPARg ligands causes inhibition of the phosphorylation The common biological effect of both SR1664 and rosiglitazone is the specic inhibition of PPARg phosphorylation at Ser273, which means that they should share a common characteristic. 14 By analyzing the binding modes of SR1664 and rosiglitazone in X-ray co-crystal structures, we found that both ligands occupied a specic hydrophobic region between helix H3 and the b3-b4 loop (Fig. 2a). Based on this structural insight, we hypothesized that the ligand binding at this specic binding site is responsible for the inhibition of PPARg phosphorylation at Ser273. To test this hypothesis, we designed two covalent inhibitors that could selectively bind at the alternate site. Considering the structures of GW9662 and SB1404, it was found that a 2-chloro-5-nitrobenzamide moiety can serve as an electrophile and covalently trap Cys313 regardless of the functional groups attached to the amide moiety ( Fig. S3 †). Therefore, we rationally designed and synthesized SB1405 (2) and SB1406 (3) containing 2-(benzyloxy)phenyl and 3-(benzyloxy)phenyl groups, respectively, instead of the phenyl group in GW9662 ( Fig. 2b and c).
On the basis of the GW9662-PPARg co-crystal structure, we expected that the additional hydrophobic benzyl moiety would occupy the specic binding site of PPARg between H3 and b3-b4, and conducted in vitro Cdk5 assays using the synthesized covalent inhibitors. Only SB1405 inhibited the phosphorylation of PPARg (Fig. 2d) without blocking the phosphorylation of the C-terminal fragment of the retinoblastoma protein (Rb peptide), a well-known Cdk5 substrate (Fig. 2e). 19 This indicated that SB1405 does not affect the fundamental kinase function of Cdk5 but blocks the phosphorylation of PPARg at Ser273 similar to the effects of rosiglitazone and SR1664. However, we did not observe this inhibitory activity in the case of SB1406, which is the structural isomer of SB1405. To explain this intriguing result, we resolved the crystal structures of the PPARg LBD complexed either with SB1405 or SB1406 to resolutions of 2.75 or 2.95Å, respectively. Similarly to SB1404, both compounds covalently bound to Cys313, but they displayed different binding modes (Fig. S5 †). In particular, the benzyl group of SB1405 occupied the specic binding site of PPARg (Fig. 2f), whereas the same moiety of SB1406 did not occupy this region (Fig. 2g). Therefore, these co-crystal structures clearly elucidated why only SB1405 inhibited PPARg phosphorylation at Ser273, demonstrating that the occupation of the hydrophobic alternate site of PPARg is essential for the inhibition of PPARg phosphorylation.
When we aligned the co-crystal structures of the SB1404-PPARg LBD and SB1405-PPARg LBD, we did not observe any considerable differences in their backbone conformations with a root-mean-square deviation (RMSD, Ca) of 0.36Å, and there was no signicant difference in the positioning of the residues around the specic binding site (Fig. S6 †), which is consistent with previous crystallography study of PPARg. 20 Based on this structural information, we assumed that the inhibition of PPARg phosphorylation is not an outcome of conformational changes but that it is probably caused by ligand-induced changes in the dynamic nature of the PPARg LBD. To test this hypothesis, we performed a hydrogen/deuterium exchange (HDX) experiment with mass spectrometry. As shown in Fig. 3 and S7, † SB1405, but not SB1404, signicantly reduced the hydrogen/deuterium exchange rate in the b-sheet compared to the ligand-free PPARg LBD. This ligand-induced reduction of the hydrogen/deuterium exchange rate at that site is an indication of change in the dynamic nature of b-sheet, and the reduced exibility of this region probably results in the subsequent inhibition of Cdk5-mediated PPARg phosphorylation at Ser273. This concept has been discussed in previous studies based on results that PPARg phosphorylation inhibitors commonly stabilized H3 and the b-sheet region. 5,14,17 However, we clearly demonstrated this concept by comparing the HDX-MS results of SB1404 and SB1405. Moreover, SB1405 did not cause any structural dynamic changes on the C-terminal indicating that SB1405 acts as a partial or non-agonist of PPARg, while anti-diabetic PPARg ligands including rosiglitazone decrease hydrogen/deuterium exchange at the C-terminal as well as the b-sheet region of PPARg in recent HDX studies. 5 Based on the mechanistic and structural understanding of PPARg phosphorylation, we selected N-(2-substituted phenyl)-2chloro-5-nitrobenzamide as a suitable molecular framework and synthesized a series of covalent inhibitors of PPARg phosphorylation containing different R groups to effectively occupy this specic binding site (Fig. 2h). As shown in Fig. 2i, these inhibitors displayed good inhibitory activities toward Cdk5mediated in vitro PPARg phosphorylation, excluding 8 which contains a hydrophilic piperazine moiety. In contrast, all these compounds exerted no inhibitory effect on the phosphorylation of the Rb peptide by Cdk5 (Fig. 2j). Taken together, the N-(2-substituted phenyl)-2-chloro-5-nitrobenz-amides possessing hydrophobic R substituents efficiently blocked Cdk5-mediated PPARg phosphorylation at Ser273, and the occupation of R substituents at the hydrophobic region between H3 and b3-b4 of PPARg appears to be sufficient for reducing the exibility of this region, which inhibits the in vitro phosphorylation of PPARg by Cdk5.

Structure-guided optimization and evaluation of covalent inhibitors for in vivo analysis
Among the N-(2-substituted phenyl)-2-chloro-5-nitrobenzamides, we selected SB1405 (2) for a further in vivo study because of its excellent inhibition potency in the in vitro Cdk5 assay (Fig. 2i) and the perfect occupancy of its benzyl moiety at the specic binding site (Fig. 2f). However, we were concerned that the in vivo evaluation of SB1405 might be limited due to its poor solubility. To solve this problem, we decided to modify the structure of SB1405 without changing the 2-chloro-5-nitrobenzamide moiety to preserve covalent anchoring with Cys313 of PPARg and a hydrophobic benzyl moiety to specically bind the alternate site for the phosphorylation inhibition. Thus, we introduced an additional moiety on the benzyl moiety of SB1405 to occupy the remaining empty hydrophobic space (brown) and the subsequent hydrophilic space (blue) (Fig. 4a). On the basis of the crystal structure of the SB1405-PPARg LBD, we designed and synthesized SB1451 (9) and SB1453 (10) containing hydrophilic piperazine moieties attached to the additional benzene rings to improve solubility (Fig. 4b). Aer synthesizing SB1451 and SB1453, we examined their inhibitory activities against PPARg phosphorylation at Ser273 in a concentration-dependent  manner and conrmed their excellent potency in the in vitro Cdk5 assay (Fig. 4c). This inhibitory potency of SB1451 and SB1453 regarding PPARg phosphorylation was also conserved in the cellular system, which was similar to that of rosiglitazone (Fig. 4e). However, they did not inhibit the basic kinase function of Cdk5 as monitored by a Western blot analysis of Rb peptide phosphorylation in vitro (Fig. 4d).

Covalent PPARg phosphorylation inhibitors exert antidiabetic effects in vivo without promoting adipogenesis
To examine the agonistic activity in PPARg-mediated transcription, we conducted two cellular assays. First, using a luciferase reporter gene assay in HEK-293T cells expressing fulllength PPARg with tandem PPAR response elements (PPRE), we conrmed that SB1451 and SB1453 had almost no classical transcriptional agonism compared with that of rosiglitazone (Fig. 5a). Secondly, we tested whether these covalent inhibitors stimulate adipogenesis by monitoring cellular lipid accumulation in 3T3-L1 cells via Oil Red O staining. 21,22 As shown in Fig. 5b, rosiglitazone (10 mM) fully stimulated adipocyte differentiation, whereas SB1451 or SB1453 (10 mM) did not trigger adipogenesis in 3T3-L1 cells, which conrms the lack of transactivation associated with these inhibitors. Although we observed a slight increase in the PPARg transcriptional activity upon treatment with SB1453 at 10 mM in the luciferase assay, this concentration was insufficient to stimulate the differentiation of pre-adipocytes into mature adipocytes.
We evaluated the anti-diabetic activity of SB1451 and SB1453 in animal models using DIO mice that are insulin-resistant with an increased level of phosphorylated PPARg at Ser273. 5 All animal experiments were performed according to procedures approved by Ulsan National Institute of Science and Technology's Institutional Animal Care and Use Committee. As shown in Fig. 5c, SB1453 effectively decreased the phosphorylation of PPARg at Ser273 in the white adipose tissue (WAT) of the DIO mice similarly to the effect of rosiglitazone. SB1451 was less potent in both the in vivo reduction of PPARg phosphorylation and the resulting anti-diabetic activity than SB1453. Previous studies clearly demonstrated that SR1664, a selective inhibitor of PPARg phosphorylation without classical agonism, exerts in vivo anti-diabetic effects and causes changes in the expression of diabetic genes that were dysregulated as a result of PPARg phosphorylation in obese animals. 12,14 Similarly to SR1664, SB1453 altered the expression of 10 out of 17 affected genes (Fig. 5e). Furthermore, we did not observe any SB1453-induced stimulation of the "agonist" gene set in the white adipose tissue of the DIO mice as dened in a previous report 14 (Fig. 5d). The glucose tolerance in the DIO mice was improved with the treatments of SB1453 at 10 mg per kg per day for 7 days, although this effect was moderate compared with that of rosiglitazone (Fig. 5f). These results indicated that SB1453 has antidiabetic actions and preferentially regulates genes sensitive to PPARg phosphorylation.
We also investigated several adverse effects, including uid retention and cardiac hypertrophy, which have been observed following treatment with glitazones. 8,23 As shown in Fig. S8a, † the treatment of rosiglitazone caused hemodilution, whereas the treatment of SB1453 had no detectable changes compared with vehicle. Furthermore, the expressions of natriuretic peptide type B (BNP), the marker gene of heart failure, or myosin heavy chain b (b-MHC), the marker gene of hypertrophy, were signicantly increased in only rosiglitazone-treated mice without changes in the heart weight ( Fig. S8b and c †). These results strongly suggest that SB1453 does not induce the adverse effects associated with the in vivo treatment of glitazones.

Covalent inhibitor selectively binds to PPARg
Lastly, we conrmed the target selectivity of SB1453 using a target identication probe. Prior to the probe design, we resolved the crystal structure of the PPARg LBD complexed with SB1453 and the SRC1 coactivator peptide (resolution, 2.30Å) to gain more information about its binding mode (Fig. 6a). As expected, SB1453 covalently bound to Cys313 on H3 and occupied the hydrophobic region between H3 and b3-b4. In addition, its piperazine moiety was positioned at the entrance of the PPARg binding pocket. Based on this structural information, we designed the target identication probe 11 containing a terminal acetylene group on the piperazine moiety to enable the uorescence labeling of target proteins via a bioorthogonal click reaction (Fig. 6b). Despite the structural change, probe 11 retained its phosphorylation inhibitory effect, indicating that this probe probably covalent bonded with Cys313 on the H3 of PPARg (Fig. 6c).
First, we conrmed the PPAR subtype selectivity of our covalent inhibitor. Probe 11 was incubated with the proteome of HEK-293T cells expressing either murine PPARa, PPARd, or PPARg, followed by a copper-catalyzed click reaction with an azide-containing Cy5 to visualize the proteins complexed with probe 11. The resulting proteome was separated by gel electrophoresis, and target proteins were visualized via uorescence scanning. Predominantly, the uorescence-labeled protein was only detected in the PPARg-transfected cells, indicating that probe 11 efficiently binds to PPARg, and not to PPARa or PPARd (Fig. S9 †). Then, we performed the same experiment with differentiated 3T3-L1 adipose cells. Interestingly, the predominant bands appeared on the uorescent gel in a dose-dependent manner without any difference in protein expression pattern ( Fig. 6d and e). These labeled proteins were identied as PPARg1 (53 kDa) and PPARg2 (57 kDa) by Western blot analysis (Fig. 6f). In fact, they are two isoforms of PPARg and have a common LBD to which SB1453 binds. This result demonstrated that our covalent inhibitor SB1453 selectively binds to target protein and might be free from general concerns regarding the nonspecic binding of irreversible covalent inhibitors. 24

Conclusions
In this study, we identied that an alternate site binding of SR1664 effectively blocks the phosphorylation of PPARg at Ser273. By comparing this binding mode with that of the conventional PPARg ligand rosiglitazone, we found that ligand binding at a specic binding site, the hydrophobic region between H3 and b3-b4, is closely related to the inhibition of PPARg phosphorylation at Ser273. To determine the functional effect of PPARg ligand binding at the specic site, we rationally designed and synthesized irreversible covalent inhibitors as chemical tools. By analysing the data from biochemical assays, X-ray crystallography, and HDX mass spectrometry with these covalent inhibitors, we found that occupation of the specic binding site in PPARg by small-molecule ligands causes the change in the dynamic nature of b-sheet and directly correlates with the inhibitory effects on the phosphorylation of PPARg at Ser273. This structural insight led us to rationally design the improved covalent inhibitors SB1451 and SB1453, which effectively inhibit the phosphorylation of PPARg at Ser273 in vitro and in adipose cells. We also demonstrated that SB1453 exerts potent anti-diabetic effects in DIO mice by blocking PPARg phosphorylation at Ser273 in the white adipose tissue without several of the adverse effects associated with glitazones. Lastly, through uorescence-based visualization of the target proteins complexed with the covalent probe 11 containing a bioorthogonal functional handle, we conrmed that our potent anti-diabetic agent SB1453 selectively binds to PPARg. This study provides a useful guideline for the structure-based rational design of SPPARgMs that effectively inhibit the phosphorylation of Ser273 for the development of anti-diabetic PPARg ligands. It can be expected that new classes of PPARgtargeting anti-diabetic drugs will be developed in accordance with this guideline.